JP5621773B2 - Classification hierarchy re-creation system, classification hierarchy re-creation method, and classification hierarchy re-creation program - Google Patents

Classification hierarchy re-creation system, classification hierarchy re-creation method, and classification hierarchy re-creation program Download PDF

Info

Publication number
JP5621773B2
JP5621773B2 JP2011521779A JP2011521779A JP5621773B2 JP 5621773 B2 JP5621773 B2 JP 5621773B2 JP 2011521779 A JP2011521779 A JP 2011521779A JP 2011521779 A JP2011521779 A JP 2011521779A JP 5621773 B2 JP5621773 B2 JP 5621773B2
Authority
JP
Japan
Prior art keywords
classification
hierarchy
re
data
group
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
JP2011521779A
Other languages
Japanese (ja)
Other versions
JPWO2011004529A1 (en
Inventor
弘紀 水口
弘紀 水口
大 久寿居
大 久寿居
Original Assignee
日本電気株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority to JP2009160071 priority Critical
Priority to JP2009160071 priority
Application filed by 日本電気株式会社 filed Critical 日本電気株式会社
Priority to PCT/JP2010/002855 priority patent/WO2011004529A1/en
Publication of JPWO2011004529A1 publication Critical patent/JPWO2011004529A1/en
Application granted granted Critical
Publication of JP5621773B2 publication Critical patent/JP5621773B2/en
Application status is Active legal-status Critical
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types

Description

  The present invention relates to a classification hierarchy re-creation system, a classification hierarchy re-creation method, and a classification hierarchy re-creation program for reconstructing a hierarchical classification and creating a new classification hierarchy.

  Patent Document 1 describes a data division method that divides multidimensional data into items having a hierarchical structure and divides the data into appropriate groups that match the purpose of analysis. When receiving the data group and the classification hierarchy of the data group, the data dividing device described in Patent Document 1 generates a classification hierarchy obtained by deleting a non-characteristic hierarchy from the classification hierarchy based on the distribution of the received data group. Output. Specifically, the attribute indicating whether or not the division target group is characteristic by performing a statistical test based on the distribution of the data group (division target group) with the specific classification as the division target. Determine. Next, the dividing unit divides the group to be divided into child group groups belonging to the child hierarchy based on the determination result, and newly sets the group to be divided. Then, the integration unit integrates the child groups that are not characteristic into the parent group based on the attribute of the determination result. Specifically, the integration unit deletes the non-characteristic hierarchy and leaves only the characteristic hierarchy. Therefore, it is possible to obtain classifications up to a characteristic child hierarchy by tracing the output classification hierarchy in order from the parent classification.

  Patent Document 2 describes a term dictionary generation method for outputting a relationship between terms based on input document data. In the term dictionary generation method described in Patent Document 2, first, related words are selected based on each word and position information of document data. Next, a graph with words and related words as nodes is created. In addition, the co-occurrence statistics are calculated for every two combinations of nodes in the graph, and the similarity is calculated from a synonym dictionary and other document data. Then, the graph is converted based on the conversion rule using the co-occurrence statistic and the similarity value.

  Patent Document 3 describes a document organizing apparatus that automatically classifies a large number of document groups stored in an information processing apparatus with high accuracy according to the characteristics of the document group. The document organizing apparatus described in Patent Document 3 defines a support sup (H → B) and a certainty factor conf (H → B) representing the co-occurrence appearance frequency of a keyword pair (H, B). Then, the XY plane defined by the point (X, Y) = (conf (kw → wi), conf (wi → kw)) is divided into five, and the hierarchical relationship, the equivalence relationship, and the associative relationship are determined.

  Patent Document 4 describes a classification system generating device that automatically constructs a hierarchical classification system from a flat classification frame. In the classification system generation device described in Patent Document 4, a cluster is generated by clustering starting from a non-hierarchical type (that is, a flat classification frame). Then, a hierarchical structure classification system is prepared using these generated clusters as the upper classification frames, and after focusing on the upper classification frames (that is, clusters) whose classification accuracy is lower than the reference value, reclustering is performed. I will extend the hierarchy. In addition, the classification system generation apparatus described in Patent Document 4 classifies the classification system of the document classification unit when the classification accuracy of the existing classification system is lower than the reference value or when the classification system is modified according to the situation. Stored in the system storage unit to be optimized. Then, the classification accuracy is improved by evaluating and changing the classification based on the classified document input from the document input unit or the sample document representing the situation.

JP 2008-299382 A (paragraphs 0027, 0047 to 0048, 0079) Japanese Patent Laid-Open No. 11-96177 (paragraphs 0015 to 0017, FIG. 1) Japanese Patent Laying-Open No. 2005-266866 (paragraphs 0021 and 0051, FIG. 4) JP 2000-10996 (paragraphs 0081, 0084 to 0085, FIG. 11)

  In the data dividing method described in Patent Document 1, a hierarchy that is not characteristic is deleted, and thus there is a problem that the hierarchy that is a deletion target cannot be classified. For example, in the data division method described in Patent Document 1, it is good if the viewpoint matching the data characteristics is in the classification hierarchy, but if there is no viewpoint matching the data characteristics, an appropriate classification hierarchy cannot be obtained. Even in a hierarchy that is not a classification target, a classification that considers the hierarchical relationship of the hierarchy or a classification that integrates classifications with the same meaning (for example, classification 1 and classification 2 are assigned to exactly the same data) Can be created as a category with the same meaning).

  In addition, the data division method described in Patent Document 1 has a problem that it is not efficient because it is necessary to make a determination for all layers in order to determine whether each layer is characteristic. Similarly, in the term dictionary generation method described in Patent Document 2, it is necessary to calculate co-occurrence statistics and similarities based on the relationships between words corresponding to all nodes in order to convert the relationships between nodes. Yes, there is a problem that it is not efficient. The document organizing apparatus described in Patent Document 3 also has a problem that it is not efficient because it generates a directory file based on all stored keywords.

  Further, the classification system generation device described in Patent Document 4 stratifies classification frames by repeating clustering of classification frames based on the degree of association with a sample document. However, since the degree of relevance is determined based on the appearance frequency of words in each cluster, the document classification apparatus described in Patent Document 4 integrates classifications that consider hierarchical relationships of hierarchies and classifications that have the same meaning. There is a problem that classification is not possible.

  Therefore, the present invention efficiently creates a classification hierarchy that considers the hierarchical relationship of classifications and a classification hierarchy that integrates classifications with the same meaning when reconstructing an existing classification hierarchy and creating a new classification hierarchy. To provide a classification hierarchy re-creation system, classification hierarchy re-creation method and classification hierarchy re-creation program

  The classification hierarchy re-creation system according to the present invention clusters a data group associated with a hierarchical classification, and extracts a classification satisfying a predetermined condition from the classifications corresponding to each data in the cluster. Based on the classification group and the co-occurrence degree, the clustering means for creating the classification group which is a group, the co-occurrence degree calculation means for calculating the co-occurrence degree of the two classifications selected from the classification group, and re-classifying the classification hierarchy. It comprises a classification hierarchy re-creating means for creating.

In the classification hierarchy re-creation method according to the present invention, the clustering means of the data processing device clusters the data group associated with the hierarchical classification, and the clustering means includes the classification corresponding to each data in the cluster, A data group is created by creating a classification group, which is a group from which classifications satisfying a predetermined condition are extracted, and the co-occurrence degree calculation means of the data processing device calculates the co-occurrence degree of two classifications selected from the classification group The classification hierarchy re-creating means re-creates the classification hierarchy based on the classification group and the co-occurrence degree.

  The classification hierarchy re-creation program according to the present invention clusters a data group associated with a hierarchical classification in a computer, and a classification satisfying a predetermined condition among the classifications corresponding to each data in the cluster Clustering process to create a classification group that is a group extracted from, a co-occurrence degree calculation process to calculate the co-occurrence degree of two classifications selected from the classification group, and based on the classification group and co-occurrence degree, A classification hierarchy re-creation process for re-creating a hierarchy is executed.

  According to the present invention, when a new classification hierarchy is created by reconstructing an existing classification hierarchy, a classification hierarchy that considers the hierarchical relationship of classifications or a classification hierarchy that integrates classifications with the same meaning is efficiently created. it can.

It is a block diagram which shows the example of the classification hierarchy recreation system in the 1st Embodiment of this invention. It is explanatory drawing which shows the example of the data group input into the input means 11, and its classification. It is a flowchart which shows the example of operation | movement of the data processor 100 in 1st Embodiment. It is explanatory drawing which shows the example of a classification hierarchy. It is explanatory drawing which shows the example of a cross tabulation table. It is explanatory drawing which shows the example of the cross tabulation table of the result of having been divided | segmented. It is explanatory drawing which shows the example of the calculation result of a co-occurrence degree. It is explanatory drawing which shows the example in the middle of updating a classification hierarchy. It is explanatory drawing which shows the example of the result of having updated the classification | category hierarchy. It is explanatory drawing which shows the example of the updated classification | category hierarchy. It is explanatory drawing which shows the example of the updated classification | category hierarchy. It is a block diagram which shows the example of the classification hierarchy recreation system in the 2nd Embodiment of this invention. It is explanatory drawing which shows the example of data with a structure. It is a block diagram which shows the example of the classification hierarchy recreation system in the 3rd Embodiment of this invention. It is a flowchart which shows the example of operation | movement of the data processor 100 in 3rd Embodiment. It is explanatory drawing which shows the example of the data group which the input means 11 receives. It is explanatory drawing which shows the example of a classification hierarchy. It is explanatory drawing which shows the example of a cross tabulation table. It is explanatory drawing which shows the example of the result of having divided | segmented the cross tabulation table. It is explanatory drawing which shows the example of a calculation result of a co-occurrence score. It is explanatory drawing which shows the example of a classification hierarchy. It is explanatory drawing which shows the example of a classification hierarchy. It is a block diagram which shows the minimum structure of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

Embodiment 1. FIG.
FIG. 1 is a block diagram showing an example of a classification hierarchy recreation system in the first exemplary embodiment of the present invention. The classification hierarchy re-creation system in this embodiment includes a data processing device 100, a data storage device 101, an input unit 11, and an output unit 16. The input unit 11 is an input device such as a keyboard, for example, but the mode of the input unit 11 is not limited to the keyboard. For example, the input unit 11 may be an input interface that receives data from another device. The output unit 16 is an output device such as a display device, for example, but the mode of the output unit 16 is not limited to the display device. For example, the output unit 16 may be an output interface that transmits data to another device.

  The data processing apparatus 100 includes clustering means 13, co-occurrence degree calculation means 14, and classification hierarchy update means 15.

  The data storage device 101 also includes a classification hierarchy storage unit 12 that stores a hierarchical relationship of classifications (hereinafter referred to as a classification hierarchy). The classification hierarchy is a hierarchy that represents the vertical relationship of classification, and is represented, for example, by a directed graph structure with classification as a node. In the following description, the case where the classification hierarchy is represented by an effective graph structure having classification as a node will be described. However, the classification hierarchy is not limited to the above structure. The classification hierarchy may be another structure that can indicate the hierarchical relationship of each classification. The classification hierarchy storage unit 12 is realized by, for example, a magnetic disk device provided in the data storage device 101. Each of the above means operates as follows.

  The input unit 11 receives the input data group and the classification of each data, and notifies the clustering unit 13 of it. FIG. 2 is an explanatory diagram showing an example of an input data group and its classification. In the example shown in FIG. 2, the data and the classification to which the data belongs (hereinafter referred to as data classification or simply “classification”) are represented by one record, and the entire table including the record is the data group. Represents. In addition, "..." in a table | surface represents omission. In the example shown in FIG. 2, a plurality of classifications separated by “,” (comma) represent classifications to which each data belongs. For example, “text data 1” in the first record indicates that it belongs to the classifications “F”, “G”, and “H”.

  The clustering means 13 receives the data group and the classification of each data from the input means 11, and clusters the received data group. The clustering means 13 may cluster the data group using a clustering method such as K-MEANs, for example. The clustering means 13 may use a method other than K-MEANs as a clustering method.

  Next, the clustering means 13 aggregates the data in each cluster for each classification, and groups the classification having a large number of data for each cluster. For example, the clustering means 13 creates a cross tabulation table using the classification corresponding to each data in each cluster. Specifically, the clustering unit 13 arranges information indicating clusters horizontally and information indicating classification vertically, and creates a cross tabulation table having values of the number of data of each cluster and classification. Then, the clustering means 13 refers to the tabulation table, marks a portion with a large number of data, and groups the marked portions for each cluster.

  Next, the clustering means 13 refers to the classification hierarchy, and divides the classification group when the marked classification group in the cluster (that is, the grouped classification) is far from the hierarchy. Then, the clustering means 13 notifies the co-occurrence degree calculating means 14 of the classification group created based on the division result (hereinafter referred to as a classification group).

  The co-occurrence degree calculating means 14 receives the classification group and calculates the co-occurrence degree for each combination of two classifications selected from the classification group. Here, co-occurrence means that two classifications appear (belong) to one data. The co-occurrence degree is a statistic calculated based on the co-occurrence and is a value indicating the degree of co-occurrence. The co-occurrence degree calculation means 14 calculates the co-occurrence degree of each classification, for example, using the number of data in which two classifications co-occur as a denominator and the number of data belonging to each classification as a numerator. For example, the number of data in which the classification “F” and the classification “G” co-occur is 10 and the number of data in the classification “G” is 9. At this time, the co-occurrence degree calculating means 14 calculates the co-occurrence degree P such that P (classification “F”, classification “G” | classification “G”) = 9/10 = 0.9. In the following description, the number of data in which two classifications co-occur is described as a co-occurrence frequency. In the above example, the co-occurrence frequency of the classification “F” and the classification “G” is 10.

  The classification hierarchy updating means 15 updates the classification hierarchy by creating the hierarchical relationship of the classifications and integrating the classifications using the classification groups and the co-occurrence degrees. First, the classification hierarchy update unit 15 extracts one classification group and extracts two classifications from the classification group. If the two extracted categories have a co-occurrence degree equal to or greater than a predetermined threshold and satisfy the inclusion relationship, the classification hierarchy update unit 15 creates a parent-child hierarchical relationship for the two categories. To do. On the other hand, if the two extracted categories have a co-occurrence degree equal to or greater than a predetermined threshold and further satisfy the agreement relationship, the classification hierarchy update unit 15 integrates the two categories. The classification hierarchy updating means 15 updates the classification hierarchy by repeating the above processing for the combination of two classifications in the group and for all the classification groups.

  Here, the inclusion relationship refers to a relationship in which one concept is wide and the other concept is narrow, and a broad concept includes a narrow concept in the concepts indicated by the two classifications. In addition, the consent relationship refers to a relationship in which both concepts are included in the same broad concept in the concepts indicated by the two classifications. That is, the classification hierarchy update unit 15 determines whether the two classifications are inclusive relations or consent relations using the co-occurrence degree, and updates the classification hierarchy based on the determined relations.

  The output means 16 outputs the contents of the updated classification hierarchy to a display device or the like.

  The clustering means 13, the co-occurrence degree calculating means 14, and the classification hierarchy updating means 15 are realized by a CPU of a computer that operates according to a program (classification hierarchy recreation program). For example, the program is stored in a storage unit (not shown) of the data processing apparatus 100, and the CPU reads the program and operates as the clustering unit 13, the co-occurrence degree calculating unit 14, and the classification hierarchy updating unit 15 according to the program. May be. Further, the clustering means 13, the co-occurrence degree calculating means 14, and the classification hierarchy updating means 15 may each be realized by dedicated hardware.

  Next, the operation will be described. FIG. 3 is a flowchart showing an example of the operation of the data processing apparatus 100 in the present embodiment.

  First, when the input unit 11 notifies the clustering unit 13 of the received data group, the clustering unit 13 performs clustering based on the data group (step S1). The clustering means 13 can use a clustering method suitable for the received data as a clustering method. For example, the clustering unit 13 may use a well-known method such as K-MEANS. In the present embodiment, the case where the clustering unit 13 clusters text data will be described, but the data group to be clustered is not limited to text data. For example, the clustering unit 13 may cluster binary data such as sound and images as a data group.

  Next, the clustering means 13 refers to the classification hierarchy stored in the classification hierarchy storage means 12, creates a cross tabulation table of each clustered cluster and data classification, and creates a classification group (step S2). FIG. 4 is an explanatory diagram illustrating an example of a classification hierarchy. FIG. 5 is an explanatory diagram showing an example of a cross tabulation table.

  The example shown in FIG. 4 indicates that the classification hierarchy is expressed by a directed graph structure in which the classification is a node. Further, the example shown in FIG. 5 indicates that the cross tabulation table is composed of a table in which information indicating clusters is arranged horizontally and information indicating classification is arranged vertically. Further, the values of the cross tabulation table illustrated in FIG. 5 indicate the number of data belonging to the classification in the data existing in the cluster (that is, the value obtained by counting the number of data belonging to each classification for the data in the cluster). But this is an example. For example, a value obtained by dividing the number of data by the total number of data in the cluster may be used, or a value obtained by dividing the number of data by the total number of data in the classification may be used.

  Here, the clustering means 13 marks cells above a certain threshold. In the example shown in FIG. 5, the marked portion is expressed by being surrounded by a thick line, and indicates that the clustering unit 13 has marked a cell having a threshold value of 10 or more. The marked part indicates that the data included in the cluster belongs to many classifications. For example, “Cluster 1” illustrated in FIG. 5 indicates that it includes a lot of data belonging to the classification H, the classification I, and the classification J. Here, the fact that there are many data belonging to the classification means that it is equal to or more than a predetermined threshold value.

  The clustering means 13 creates a classification group based on the classification marked for each cluster. For example, in the example illustrated in FIG. 5, the clustering unit 13 groups the classifications (classification H, classification I, and classification J) marked in “cluster 2” into one group (classification group). Next, the clustering means 13 refers to the cross tabulation table and the classification hierarchy, and divides the classification group having a large hierarchical distance (step S3). The clustering means 13 determines whether or not the hierarchical distance is greater than or equal to the threshold for each classification of the classification group. If the hierarchical distance is equal to or greater than the threshold, the clustering unit 13 divides the classification group. Here, the hierarchical distance is an index indicating the degree of separation between the hierarchized classifications, and in the present embodiment, means the shortest hop number in the classification hierarchy of the two classifications.

  Hereinafter, a method in which the clustering unit 13 divides the classification group when the threshold is 5 hops will be described with reference to FIGS. 4 and 5. In the example shown in FIGS. 4 and 5, in the classification group of “cluster 3” (classification O, classification P, classification Q, classification R), classification O and classification Q, classification O and classification R, classification P and classification Q, and Since the classification P and the classification R are 6 hops away from each other, they become division targets. The above-mentioned classification pairs are divided into separate groups (classification O, classification P) and (classification Q, classification R). An example of the result of dividing the cross tabulation table is shown in FIG. In the example illustrated in FIG. 6, the classification group of “cluster 3” (classification O, classification P, classification Q, classification R) is classified into the classification group of “cluster 3” (classification O, classification P) and “cluster 3 ′”. It shows that it was divided into the following classification groups (classification Q, classification R). In the following description, the cluster numbers illustrated in FIG. 6 are described as classification group numbers (hereinafter referred to as group numbers).

  Next, the co-occurrence degree calculation means 14 calculates the co-occurrence degree of two classifications selected from the classification group (step S4). FIG. 7 is an explanatory diagram illustrating an example of a calculation result of the co-occurrence degree. The table illustrated in FIG. 7 includes “class 1” and “class 2” which are two classifications for which the group number and the degree of co-occurrence are calculated, and “co-occurrence score 1” indicating the degree of co-occurrence of each class. ”And“ Co-occurrence score 2 ”. In the following description, “co-occurrence score 1” and “co-occurrence score 2” indicating the degree of co-occurrence are conditional probabilities of co-occurrence of “class 1” and “class 2”, respectively. That is, “co-occurrence score 1” is the probability of co-occurrence for “class 1”, and “co-occurrence score 2” is the probability of co-occurrence for “class 2”. The value of “co-occurrence score 1” can be calculated by the following (formula 1), and the value of “co-occurrence score 2” can be calculated by the following (formula 2).

  Co-occurrence score 1 = P (Class 1, Class 2 | Category 1) = Co-occurrence frequency of Class 1 and Class 2 / Frequency of Class 1 (Formula 1)

  Co-occurrence score 2 = P (Category 1, Category 2 | Category 2) = Co-occurrence frequency of Category 1 and Category 2 / Frequency of Category 2 (Formula 2)

  The co-occurrence degree calculation means 14 determines whether the two classifications are inclusion relations or consent relations based on these two values (ie, co-occurrence score 1 and co-occurrence score 2).

  For example, when one of the co-occurrence score 1 and the co-occurrence score 2 is high, it can be said that there is an inclusion relationship between the classification corresponding to the high score and the other classification. Moreover, when the scores of both the co-occurrence score 1 and the co-occurrence score 2 are high, it can be said that there is an agreement relationship between the two categories. This is because the common part that is the numerator is the same, but the classification frequency that is the denominator is different.

  The case where the co-occurrence score 1 is high and the co-occurrence score 2 is low will be specifically described as an example. When the co-occurrence score 1 is high, almost all data belonging to the category 1 also belongs to the category 2. In other words, when the co-occurrence score 2 is small, the data belonging to the category 2 belongs to various data other than the category 1. Therefore, it can be said that category 2 is larger than category 1, and category 2 includes category 1. Conversely, when the co-occurrence score 2 is high and the co-occurrence score 1 is low, it can be said that the classification 1 includes the classification 2.

  On the other hand, if two co-occurrence scores (ie, co-occurrence score 1 and co-occurrence score 2) are the same, the same data often appears in each category (ie, category 1 and category 2). It can be said that classification 1 and classification 2 are agreements.

  Next, the classification hierarchy updating means 15 updates the classification hierarchy based on the classification group and the co-occurrence degree (step S5). As a result of the determination based on the co-occurrence degree, the classification hierarchy updating unit 15 updates the two classifications as parent-child classifications when the relation between the two classifications satisfies the inclusion relation. On the other hand, when the relationship between the two categories satisfies the agreement relationship, the category hierarchy updating unit 15 integrates the two categories into one category. The classification hierarchy update unit 15 determines the level of the co-occurrence score using a threshold value. Hereinafter, this threshold is referred to as a co-occurrence score threshold.

  Hereinafter, the process of updating the classification hierarchy will be described using the examples shown in FIGS. 4 and 7. Here, it is assumed that the co-occurrence score threshold is preset in the system. The classification hierarchy updating unit 15 determines that the co-occurrence score threshold is high when the co-occurrence score threshold is 0.7 or more, and determines that the co-occurrence score threshold is low when the co-occurrence score threshold is 0.3 or less.

  According to the co-occurrence score of classification G and classification H of “group 1” illustrated in FIG. 7, it can be said that “co-occurrence score 1” is high and “co-occurrence score 2” is low. Therefore, it can be seen that these two classifications have an inclusion relationship, and classification H has a parent relationship and classification G has a child relationship. Therefore, the classification hierarchy updating unit 15 updates the classification hierarchy so that the classification H illustrated in FIG. 4 has a parent and the classification G has a child relationship. An example in the middle of updating the classification hierarchy is shown in FIG. In the example shown in FIG. 8, it can be seen that the category G is updated as a child of the category H. The broken line drawn from the category B to the category G is a line indicating the parent-child relationship before update. The classification hierarchy update unit 15 may or may not delete the parent-child relationship before the update. In the following description, the parent-child relationship before update will be deleted later.

  In addition, when viewing the co-occurrence scores of classification H and classification I of “group 2” illustrated in FIG. 7, it can be said that “co-occurrence score 2” is high and “co-occurrence score 1” is low. Therefore, it can be seen that these two categories also have an inclusive relationship, with category H having a parent and category I having a child relationship. Similarly, from the co-occurrence score of classification H and classification J, it can be seen that classification H has a parent relationship and classification J has a child relationship. On the other hand, it can be seen that the classification I and the classification J are in agreement because both co-occurrence scores are high. Therefore, the classification hierarchy update unit 15 integrates these two classifications.

  FIG. 9 shows an example of the result of updating the classification hierarchy based on the classification group “group 2”. The classification hierarchy illustrated in FIG. 9 is updated by “Group 1” and “Group 2”. In addition, when integrating consent-related categories, the parent categories of the categories may be different. In this case, the classification hierarchy updating unit 15 integrates a classification with a small amount of data into a classification with a large amount of data included in the two classifications to create one classification.

  In addition, since it is understood that the classification O and the classification P of “Group 3” illustrated in FIG. 7 are in a consent relationship, the classification hierarchy updating unit 15 integrates these two classifications. On the other hand, the classification Q and the classification R of the “group 3 ′” illustrated in FIG. 7 are neither inclusive nor in agreement, so the classification hierarchy updating unit 15 does not update the classification hierarchy.

  FIG. 10 shows an example of the classification hierarchy updated as a result of the above. Here, the classification surrounded by the thick line illustrated in FIG. 10 is a classification in which the data to which it belongs. The classification hierarchy update unit 15 may or may not delete the parent-child relationship before the update (relationship connected by a broken line in the figure). In the case of leaving without deleting, for example, a request to classify data using the classification hierarchy before update can be satisfied.

  Furthermore, the classification hierarchy update unit 15 may perform processing for a classification for which no data belongs. For example, the classification hierarchy update unit 15 may delete a classification when a classification with no data belonging to does not have a child classification. For example, in the example shown in FIG. 10, since there is no data belonging to the classification L, the classification M, and the classification N, the classification hierarchy updating unit 15 may delete these classifications.

  Further, the classification hierarchy updating means 15 deletes the classification for a classification having no data belonging to the classification and having only one child classification, and determines the parent classification and the child classification of the classification to be deleted. A hierarchical relationship may be created between them. That is, the classification hierarchy update unit 15 may create a hierarchical relationship in which the grandchild classification is a child classification. This is because there is not much point in keeping such a hierarchy of classifications that has only one child classification. For example, since the class E has only the class O + P, the class hierarchy updating unit 15 deletes the class E and directly creates a vertical relationship with the class B and the class O + P. As a result of the above, an example of the updated classification hierarchy is shown in FIG.

  As described above, according to the present embodiment, the clustering means 13 clusters the data group associated with the hierarchical classification. Then, the clustering unit 13 creates a classification group in which classifications satisfying a predetermined condition (for example, a condition “the number of belonging data is large”) among classifications corresponding to each data in the cluster are created. Then, when the co-occurrence degree calculating unit 14 calculates the co-occurrence degree of two classifications selected from the classification group, the classification hierarchy updating unit 15 recreates the classification hierarchy based on the classification group and the co-occurrence degree. Therefore, when a new classification hierarchy is created by reconstructing an existing classification hierarchy, it is possible to efficiently create a classification hierarchy that considers the hierarchical relationship of classifications, or a classification hierarchy that integrates classifications having the same meaning.

  That is, according to the present embodiment, the classification hierarchy updating unit 15 creates the hierarchical relationship of the classification and integrates the classification based on the co-occurrence degree of the classification in the classification group. Create hierarchical relationships and integrate classifications. In addition, according to the present embodiment, the clustering unit 13 creates a group of similar classifications in advance, and the co-occurrence degree calculating unit 14 calculates only the co-occurrence degree in the group. The classification hierarchy can be updated effectively in consideration.

Embodiment 2. FIG.
FIG. 12 is a block diagram illustrating an example of a classification hierarchy recreation system according to the second exemplary embodiment of the present invention. The second embodiment is different from the first embodiment in that the input unit 11 is changed to the second input unit 21 and the clustering unit 13 is changed to the second clustering unit 23. In addition, about the structure similar to 1st Embodiment, the code | symbol same as FIG. 1 is attached | subjected and description is abbreviate | omitted.

  The classification hierarchy re-creation system in this embodiment includes a data processing device 100, a data storage device 101, a second input unit 21, and an output unit 16. The data storage device 101 is the same as that in the first embodiment, and the mode of the second input means 21 is the same as that of the input means 11 in the first embodiment. The second input means 21 receives the input structured data group and the classification of each data. In the following description, structured data means data to which a name for identifying each portion of structured data (hereinafter referred to as a structure portion name) is assigned.

  FIG. 13 is an explanatory diagram showing an example of structured data. FIG. 13 shows an example of patent data. Patent data has structural information such as summary, purpose, and problem in advance. The second input means 21 receives such structured data as one data. In the above description, the case where the second input unit 21 receives text data as structured data has been described. However, the second input unit 21 may receive audio data, image data, or the like. In the case of voice data, the structured data may be a speech part of a specific speaker of voice, and in the case of image data, the structured data may be a specific person of the image.

  Further, the second input means 21 also receives a structure part name that is to be analyzed (clustered object) by the second clustering means 23 described later. The structure part name can be said to be the name of structure information. In the example illustrated in FIG. 13, the structure part name is a summary, a purpose, a problem, or the like. The second input means 21 may receive a plurality of structure part names. For example, the second input means 21 may receive two structural part names of “issue” and “object of invention”.

  The data processing apparatus 100 includes second clustering means 23, co-occurrence degree calculating means 14, and classification hierarchy updating means 15. Since the co-occurrence degree calculating unit 14 and the classification hierarchy updating unit 15 are the same as those in the first embodiment, the description thereof is omitted.

  The second clustering means 23 receives the structured data group, the classification of each data, and the structure part name from the second input means 21, and clusters the structured data group. Specifically, the second clustering means 23 does not perform clustering based on the entire structured data, but extracts only the part corresponding to the received structure part name from each data, and extracts the information of the extracted part. Clustering is performed originally. For example, the second clustering means 23 extracts the text corresponding to the “issue” and the “purpose of the invention” from the structured data having the structure illustrated in FIG. 13, and uses only the text of this part. Clustering is performed by determining the degree of similarity. For example, the second clustering unit 23 may cluster the data group using a clustering method such as K-MEANs. Note that the second clustering means 23 may use a method other than K-MEANs as a clustering method.

  When the structured data is voice data and a specific speaker name is received as the structure part name, the second clustering means 23 extracts, for example, the waveform of the part corresponding to the speaker name, Clustering may be performed by calculating similarity. Also, in the case where the structured data is image data and a specific person name is received as the structure part name, the second clustering means 23 extracts only the area of the image in which this person is shown, and the similarity is obtained. Clustering may be performed by calculation.

  The second clustering means 23, the co-occurrence degree calculating means 14, and the classification hierarchy updating means 15 are realized by a CPU of a computer that operates according to a program (classification hierarchy recreation program). Further, the second clustering means 23, the co-occurrence degree calculating means 14, and the classification hierarchy updating means 15 may each be realized by dedicated hardware.

  Next, the operation will be described. The operation of the data processing apparatus 100 in this embodiment is the same as the flowchart illustrated in FIG. In the second embodiment, the second clustering unit 23 receives the structured data group, the classification of each data, and the structure part name from the second input unit 21 and performs clustering of the structured data group. This is the same as the first embodiment. Specifically, in the first embodiment, the clustering unit 13 performs clustering based on the entire data. On the other hand, in the second embodiment, the second clustering means 23 extracts only the part corresponding to the received structure part name from each data, and performs clustering based on the extracted part information. Other operations are the same as those in the first embodiment.

  As described above, according to the present embodiment, the second clustering unit 23 uses the data extracted from the structured data based on the structured data and the structured part name. Cluster data groups. Therefore, in addition to the effects of the first embodiment, the classification hierarchy can be recreated from the viewpoint that the user wants to analyze.

  That is, according to the present embodiment, the second clustering means 23 extracts and clusters only the part to be analyzed. Specifically, clustering is performed using the structured data and the structure part name to be analyzed. Therefore, the classification hierarchy can be updated from the viewpoint that the user wants to analyze. As described above, since the classification group can be changed by changing the analysis target, the characteristics indicated by the analysis target portion can be reflected in the classification hierarchy. For example, if the target data is patent data, the classification hierarchy can be updated from the viewpoint of wanting to divide by purpose or divide by issue.

Embodiment 3. FIG.
FIG. 14 is a block diagram illustrating an example of a classification hierarchy recreation system according to the third exemplary embodiment of the present invention. The third embodiment is different from the first embodiment in that the data processing apparatus 100 includes a re-updating unit 31. In addition, about the structure similar to 1st Embodiment, the code | symbol same as FIG. 1 is attached | subjected and description is abbreviate | omitted. That is, the data processing apparatus 100 according to the third embodiment includes clustering means 13, co-occurrence degree calculating means 14, classification hierarchy updating means 15, and re-update means 31. The clustering unit 13, the co-occurrence degree calculating unit 14, and the classification hierarchy updating unit 15 are the same as those in the first embodiment, and thus description thereof is omitted.

  The re-updating means 31 receives the updated classification hierarchy from the classification hierarchy update means 15 and instructs to re-update the classification hierarchy when the received classification hierarchy does not satisfy a predetermined condition. Here, the predetermined condition is at least one of the classification number and depth of the classification hierarchy, the number of re-updates, the presence / absence of a stop instruction from the user, or a combination thereof, but the predetermined condition includes these contents. It is not limited.

  Specifically, the re-updating unit 31 rewrites the data group classification and classification hierarchy with the updated classification hierarchy. Further, the re-updating unit 31 changes the threshold for clustering and the threshold for determining the inclusion relationship and the consent relationship (that is, the co-occurrence score threshold) by the classification hierarchy updating unit 15 to a relaxed value. Then, the re-updating unit 31 instructs the clustering unit 13 to re-create the classification hierarchy.

  The clustering means 13, the co-occurrence degree calculating means 14, the classification hierarchy updating means 15, and the re-update means 31 are realized by a CPU of a computer that operates according to a program (classification hierarchy re-creation program). The clustering unit 13, the co-occurrence degree calculating unit 14, the classification hierarchy updating unit 15, and the re-updating unit 31 may be realized by dedicated hardware.

  Next, the operation will be described. FIG. 15 is a flowchart showing an example of the operation of the data processing apparatus 100 in the present embodiment. The processing until the input means 11 receives data and the classification hierarchy update means updates the classification hierarchy is the same as the processing in steps S1 to S5 in FIG. The re-updating means 31 receives the updated classification hierarchy from the classification hierarchy update means 15, and determines whether or not the received classification hierarchy satisfies a predetermined condition (step S6). If the predetermined condition is not satisfied (NO in step S6), the re-updating means 31 changes the threshold value for clustering or the co-occurrence score threshold value to a relaxed value (step S7), and recreates the classification hierarchy. The clustering means 13 is instructed to do so. Thereafter, the processes of steps S1 to S6 are repeated. On the other hand, when the predetermined condition is satisfied (YES in step S6), the re-updating unit 31 ends the update process.

  As described above, according to the present embodiment, the re-updating unit 31 instructs to update again the classification hierarchy recreated by the classification hierarchy updating unit 15. Specifically, the re-updating unit 31 determines the condition for creating the classification group and the co-occurrence degree for re-creating the classification hierarchy when the re-created classification hierarchy does not satisfy the predetermined requirement. Change the conditions. Then, the clustering means 13 creates a classification group from which the classification satisfying the changed condition is extracted, and the classification hierarchy recreating means 15 recreates the classification hierarchy based on the changed condition. Therefore, in addition to the effects of the first embodiment, a classification hierarchy closer to the condition can be obtained. That is, even if the conditions are not met, the re-updating unit 31 performs the update again, whereby a classification hierarchy closer to the conditions can be obtained.

  Hereinafter, the present invention will be described with reference to specific examples, but the scope of the present invention is not limited to the contents described below. In this embodiment, a specific example will be described based on the block diagram illustrated in FIG. 1 and the flowchart illustrated in FIG.

  First, when the input unit 11 notifies the clustering unit 13 of the received data group, the clustering unit 13 performs clustering based on the data group (step S1 in FIG. 3). An example of a data group received by the input means 11 is shown in FIG. The data group illustrated in FIG. 16 includes “data” and “classification” in one record. In the present embodiment, text data is described as an example of data, but the data may be voice or image. In addition, the classifications illustrated in FIG. 16 are separated by commas and indicate that a plurality of classifications are specified.

  Hereinafter, a case where the clustering means 13 clusters this data will be described. The clustering means 13 performs clustering using a clustering method suitable for data. In the case of the present embodiment, since the received data is text data, the clustering means 13 uses the K-MEANS method for calculating the similarity using the text of each data as vector data. Specifically, the clustering means 13 first morphologically analyzes the text of each data and divides it into words. Next, the clustering means 13 converts the data into vector data in which the dimension is a word and the value is the number of words. Next, the clustering means 13 creates K clusters from the cosine similarity between the vector data. In this embodiment, it is assumed that K = 4, and the clustering means 13 creates four clusters.

  Note that when the received data is not text data but binary data such as voice or image, the clustering means 13 may use a method suitable for each data. For example, in the case of voice data, the clustering unit 13 may read the voice waveform data and calculate and cluster based on the similarity. In the case of an image, a color histogram may be generated from the image, and calculation may be performed based on the similarity to perform clustering.

  Next, the clustering means 13 refers to the classification hierarchy stored in the classification hierarchy storage means 12, creates a clustering result cluster and classification cross tabulation table, and creates a classification group (step S2 in FIG. 3). An example of the classification hierarchy is shown in FIG. 17, and an example of the cross tabulation table is shown in FIG.

  The classification hierarchy illustrated in FIG. 17 has a directed graph structure with classification as a node. In the example shown in FIG. 17, “main category” is the root classification, classifications “society” and “nature” exist in the lower hierarchy of the classification, and various broad classifications exist in the lower hierarchy of the classification “society”. Indicates that

  In addition, the cross tabulation table illustrated in FIG. 18 is a table in which information indicating clusters is arranged horizontally and information indicating classification is arranged vertically. The values in the cross tabulation table illustrated in FIG. 18 indicate the number of data belonging to each category, which is data existing in the cluster. However, the value illustrated in FIG. 18 is an example, and the value may be a value obtained by dividing the number of data by the total number of data in the cluster, or may be a value obtained by dividing the number of data by the total number of data for classification. In the present embodiment, it is assumed that only data belonging to a category of “Society” or lower is input.

  Here, the clustering means 13 marks cells above a certain threshold. In the example shown in FIG. 18, the marked portion is expressed by being surrounded by a thick line, and indicates that the clustering unit 13 has marked a cell having a threshold value of 10 or more. The marked part indicates that the data included in the cluster belongs to many classifications. For example, “Cluster 1” illustrated in FIG. 18 includes a lot of data belonging to the classification “Transplant” and the classification “Relatives”. Here, the fact that there are many data belonging to the classification means that it is equal to or more than a predetermined threshold value.

  The clustering means 13 creates a classification group based on the classification marked for each cluster. For example, in the example shown in FIG. 18, the clustering means 13 classifies the classifications (“Transplant”, “Relatives”) marked in “Cluster 1” into one group (classification group). In addition, the clustering means 13 selects a group (“health”, “medicine”, “transplant”) from “cluster 2” and a group “administration”, “diplomat” from “cluster 3”. ) Groups (“Home”, “Child-raising”) are created from “Cluster 4”.

  Next, the clustering unit 13 refers to the cross tabulation table and the classification hierarchy, and divides the classification group having a hierarchical distance (step S3 in FIG. 3). The clustering means 13 determines whether or not the hierarchical distance is greater than or equal to the threshold for each classification of the classification group. If the hierarchical distance is equal to or greater than the threshold, the clustering unit 13 divides the classification group. In this embodiment, the hierarchical distance means the shortest number of hops in the classification hierarchy of two classifications.

  Hereinafter, the case where the threshold is 5 hops will be described with reference to FIG. In the example shown in FIG. 17, in the group (“Transplant”, “Relative”), “Transplant” and “Category” are 5 hops away from each other, so that they are to be divided. Therefore, this group is divided into (“Transplant”) and (“Relative”). An example of the result of dividing the cross tabulation table is shown in FIG. In the example shown in FIG. 19, it can be seen that the classifications “transplant” and “relative” of “cluster 1” are divided into “cluster 1” and “cluster 1 ′”, respectively. In the following description, the cluster numbers illustrated in FIG. 19 are described as group numbers.

  Next, the co-occurrence degree calculating means 14 calculates the co-occurrence degree of two classifications selected from the classification group (step S4 in FIG. 3). Here, the co-occurrence degree is a statistic based on the co-occurrence frequencies of the two classifications. FIG. 20 shows an example of the calculation result of the co-occurrence score. The table illustrated in FIG. 20 includes a classification group number, two classifications for which the co-occurrence degree is calculated, “class 1” and “class 2”, and a “co-occurrence score” indicating the co-occurrence degree of each classification. 1 ”and“ Co-occurrence score 2 ”. In this embodiment, “co-occurrence score 1” and “co-occurrence score 2” indicating the degree of co-occurrence are conditional probabilities of co-occurring “class 1” and “class 2”, respectively. That is, “co-occurrence score 1” is the probability of co-occurrence for “class 1”, and “co-occurrence score 2” is the probability of co-occurrence for “class 2”. The value of “co-occurrence score 1” and the value of “co-occurrence score 2” can be calculated by the above (formula 1) and (formula 2), respectively.

  Specifically, the value of the co-occurrence score is calculated as follows. In “classification group 1” and “classification group 1 ′”, there is only one classification with a mark (that is, a classification to which data of a certain threshold or more belongs). Therefore, the co-occurrence degree calculation means 14 does not calculate the co-occurrence score. On the other hand, in “classification group 1” and “classification group 1 ′”, there are two classifications with marks (that is, classification “health” and “medicine”). Therefore, the co-occurrence degree calculation means 14 calculates the co-occurrence score for the two classifications “health” and “medicine” of “classification group 2” as follows.

  Here, the number of “health” and “medicine” allocated to the same data (ie, the co-occurrence frequency of “health” and “medicine”) is 16, the appearance frequency of “health” is 21, and “medicine” The appearance frequency of is assumed to be 20. At this time, each co-occurrence score is calculated as follows.

  Co-occurrence score 1 = P (health, medicine | health) = co-occurrence frequency of “health” and “medicine” / frequency of “health” = 16/21 = 0.77

  Co-occurrence score 2 = P (health, medicine | medicine) = co-occurrence frequency of “health” and “medicine” / frequency of “medicine” = 16/20 = 0.8

  Since other co-occurrence scores are calculated in the same manner, description thereof is omitted.

  Next, the classification hierarchy update unit 15 updates the classification hierarchy based on the classification group and the co-occurrence degree (step S5 in FIG. 3). The classification hierarchy update unit 15 determines the level of the co-occurrence degree (that is, the co-occurrence score) using the co-occurrence score threshold. In this embodiment, the classification hierarchy updating unit 15 determines that the co-occurrence score is high when the co-occurrence score threshold is 0.7 or more, and determines that the co-occurrence score is low when the co-occurrence score threshold is 0.2 or less. It shall be.

  According to the co-occurrence degree (co-occurrence score) of “health” and “medicine” of “group 2” illustrated in FIG. 20, it is determined that “co-occurrence score 1” is high and “co-occurrence score 2” is also high. The Therefore, it can be said that these two classifications have a consensus relationship. Further, as described above, since the appearance frequency of “health” is 21 and the appearance frequency of “medicine” is 20, it can be said that “health” is a larger classification. Therefore, the classification hierarchy updating means 15 updates the classification hierarchy by integrating “medicine” with “health”.

  On the other hand, the co-occurrence of “health” and “transplant” in “Group 2” and the co-occurrence of “medicine” and “transplant” in “Group 2” illustrated in FIG. 20 are not high. I can't say it's too low. Therefore, the classification hierarchy update unit 15 does not update the classification hierarchy.

  Further, according to the co-occurrence degree of “administration” and “diplomat” of “group 3” illustrated in FIG. 20, it is determined that “co-occurrence score 1” is low and “co-occurrence score 2” is high. Therefore, it can be said that these two categories have an inclusive relationship. Therefore, the classification hierarchy updating means 15 updates the classification hierarchy with “administration” as the parent and “diplomat” as the child.

  Similarly, the “co-occurrence score 1” and “co-occurrence score 2” of “Group 4” illustrated in FIG. 20 are determined to be high. Therefore, it can be said that these two classifications have a consensus relationship. Here, when “home” is a larger classification, the classification hierarchy updating unit 15 updates the classification hierarchy by integrating “childcare” into “home”.

  An example of the classification hierarchy obtained as a result is shown in FIG. The broken line shown in FIG. 21 is a line indicating the parent-child relationship before the classification hierarchy is updated. Further, among the classes illustrated in FIG. 21, a class in which data belonging to the class exists is represented by being surrounded by a thick line, and a class in which no data exists in the class is represented without being surrounded by a thick line. Note that the parent-child relationship before the update may or may not be deleted. In this embodiment, the classification hierarchy update means 15 will delete later.

  Furthermore, the classification hierarchy update unit 15 may perform processing for a classification for which no data belongs. In the present embodiment, a classification having no data belonging thereto and having no child classification is deleted. For example, among the categories illustrated in FIG. 21, “family law”, “diplomatic history”, and “government” are categories that do not have data belonging to the category and do not have child categories. Therefore, the classification hierarchy update means 15 updates the classification hierarchy by deleting these classifications. Further, the classification hierarchy update means 15 may delete a classification for a classification having no data belonging to the classification and has only one child classification and create a direct hierarchical relationship by moving up the child classification. Good. However, in this embodiment, since there is no such classification, the classification hierarchy is not updated. An example of the classification hierarchy obtained as a result is shown in FIG.

  In addition, when displaying the information search result, the present invention can be applied to the usage of classifying and displaying the search result. The present invention can also be applied to the case where related words defined based on the relationship between the updated classification hierarchy and the words in the classification are displayed.

  Next, the minimum configuration of the present invention will be described. FIG. 23 is a block diagram showing the minimum configuration of the present invention. The classification hierarchy re-creation system according to the present invention clusters a data group associated with a hierarchical classification, and among classifications corresponding to each data in the cluster, classifications that satisfy a predetermined condition (for example, Clustering means 81 (for example, clustering means 13) for creating a classification group (for example, classification group, classification group) that is a group from which a classification having a large number of data belonging) is extracted, and co-occurrence of two classifications selected from the classification group The degree of co-occurrence (for example, the co-occurrence degree calculating means 14) (for example, the co-occurrence degree calculating means 14) for calculating the degree (for example, calculating according to (Expression 1) and (Expression 2)), and classification based on the classification group and the co-occurrence degree Classification hierarchy re-creating means 83 (classification hierarchy updating means 15) for re-creating the hierarchy (for example, classification hierarchy).

  With such a configuration, when a new classification hierarchy is created by reconstructing an existing classification hierarchy, a classification hierarchy that considers the hierarchical relationship of classifications and a classification hierarchy that integrates classifications with the same meaning are efficiently created. it can.

  Moreover, it can be said that at least the classification hierarchy re-creation system as described below is described in any of the embodiments described above.

(1) A group of data associated with a hierarchical classification is clustered, and among classifications corresponding to each data in the cluster, a classification that satisfies a predetermined condition (for example, a classification having a large number of belonging data) Clustering means (for example, clustering means 13) for creating a classification group (for example, classification group, classification group), which is a group extracted from, and the co-occurrence degree of two classifications selected from the classification group (for example, ( Based on the co-occurrence degree calculating means (e.g., co-occurrence degree calculating means 14) calculated by Equation (1) and (Equation 2), and the classification hierarchy (for example, classification hierarchy) based on the classification group and the co-occurrence degree. A classification hierarchy recreation system comprising classification hierarchy recreation means (classification hierarchy update means 15) for recreation.

(2) A classification hierarchy recreating system in which the clustering means creates a classification group obtained by dividing the classification group (for example, classification group) when the classification in the generated classification group is more than a predetermined distance. .

(3) The co-occurrence degree calculation means calculates the co-occurrence degree based on the co-occurrence frequency that is the number of data in which two classifications co-occur and the number of data belonging to each classification, and class hierarchy re-creation means Is based on the co-occurrence degree, and the classification hierarchy is recreated based on the judgment result indicating whether the two classifications are inclusion relations or consent relations. Creation system.

(4) Classification hierarchy re-creating means, when the relationship between two classifications is an inclusion relationship, adding a hierarchy in which the inclusion classification is the parent classification and the inclusion classification is the child classification If the relationship between the two classifications is a consensus relationship, the classification hierarchy is created by creating a classification that integrates a small number of classifications with respect to a classification with a large amount of data included in the two classifications. Classification hierarchy re-creation system to re-create.

(5) When the classification hierarchy recreating means adds a hierarchy in which the included classification is a child classification, the classification hierarchy is deleted by deleting the parent-child relationship of the child classification before the classification hierarchy is recreated. Classification hierarchy re-creation system to re-create.

(6) The classification hierarchy recreating means recreates the classification hierarchy by deleting the classification when the classification with no belonging data is a classification having no child classification, In the case of a category having only one category, the category hierarchy is recreated by deleting the category and creating a hierarchical relationship between the parent category and the child category of the deleted category. Recreation system.

(7) Based on the structured data that is structured data and the structure part name that is a name for identifying each part of the structured data, the clustering unit (for example, the second clustering unit 23) A classification hierarchy re-creation system that clusters structured data groups using data extracted from structured data for the part corresponding to the name.

(8) A re-updating unit (for example, re-updating unit 31) for instructing re-updating of the re-created classification layer is provided by the classification layer re-creating unit, and the re-updating unit determines the re-created classification layer If at least one of the conditions for creating a classification group and the condition for co-occurrence for recreating a classification hierarchy is changed, the clustering means A classification hierarchy re-creation system in which a classification group is created by extracting the satisfying classifications, and the classification hierarchy re-creation means re-creates the classification hierarchy based on the changed condition.

(9) When the re-updating means does not satisfy a predetermined requirement, at least one of the number of classifications in the classification hierarchy, the depth of the classification hierarchy, the number of re-updates of the classification hierarchy, and whether or not there is a stop instruction, Classification hierarchy re-creation system that changes conditions.

(10) A classification hierarchy recreating system in which the clustering means extracts a classification having a number of data belonging to the classification larger than a predetermined number from the classification corresponding to each data in the cluster to create a classification group.

  Although the present invention has been described with reference to the embodiments and examples, the present invention is not limited to the above embodiments and examples. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  This application claims the priority on the basis of the JP Patent application 2009-160071 for which it applied on July 6, 2009, and takes in those the indications of all here.

  The present invention is preferably applied to a classification hierarchy recreating system that reconstructs a hierarchical classification and creates a new classification hierarchy.

DESCRIPTION OF SYMBOLS 11 Input means 12 Classification hierarchy storage means 13 Clustering means 14 Co-occurrence degree calculation means 15 Classification hierarchy update means 21 Second input means 23 Second clustering means 31 Reupdate means 100 Data processing device 101 Data storage device

Claims (22)

  1. Clustering means for clustering data groups associated with hierarchized classifications and creating a classification group that is a group in which classifications satisfying a predetermined condition are extracted from the classifications corresponding to each data in the cluster When,
    A co-occurrence degree calculating means for calculating the co-occurrence degree of two classifications selected from the classification group;
    A classification hierarchy recreating system comprising: a classification hierarchy recreating means for recreating the classification hierarchy based on the classification group and the co-occurrence degree.
  2. The classification hierarchy recreation system according to claim 1, wherein the clustering means creates a classification group obtained by dividing the classification group when the classification in the created classification group is more than a predetermined distance.
  3. The co-occurrence degree calculating means calculates the co-occurrence degree based on the co-occurrence frequency that is the number of data in which two classifications co-occur and the number of data belonging to each classification,
    The classification hierarchy recreating means determines whether the two classifications are inclusive relations or consent relations based on the co-occurrence degree, and based on the determination result indicating whether the two classifications are inclusion relations or consent relations, The classification hierarchy re-creation system according to claim 1 or 2, wherein the hierarchy is re-created.
  4. The classification hierarchy re-creating means adds the hierarchy of the classification by adding the hierarchy in which the inclusion classification is the parent classification and the inclusion classification is the child classification when the relationship of the two classifications is an inclusion relation. If the relationship between the two categories is a consensus relationship, the category hierarchy is recreated by creating a category that integrates a small number of categories into a category with a large amount of data. The classification hierarchy re-creation system according to claim 3.
  5. The classification hierarchy re-creation means recreates the classification hierarchy by deleting the parent-child relationship of the child classification before re-creating the classification hierarchy when adding a hierarchy with the included classification as a child classification The classification hierarchy re-creation system according to claim 4.
  6. The classification hierarchy re-creating means re-creates the classification hierarchy by deleting the classification when the classification with no belonging data is a classification with no child classification, and is a classification with no data to which the classification is 1 In the case of a class having only one class, the class is deleted, and the class hierarchy is recreated by creating a hierarchical relationship between the parent class of the class to be deleted and the child class. The classification hierarchy recreating system according to claim 5.
  7. The clustering means, based on the structured data that is structured data and the structural part name that is a name for identifying each part of the structured data, extracts a part corresponding to the structural part name from the structured data. The classification hierarchy re-creation system according to claim 1, wherein the structured data group is clustered using the extracted data.
  8. Re-updating means for instructing to re-update the classification hierarchy re-created by the classification hierarchy re-creation means,
    When the re-created classification hierarchy does not satisfy a predetermined requirement, the re-update means has at least one of a condition for creating a classification group and a condition for co-occurrence for re-creating a classification hierarchy. Change one condition,
    The clustering means creates a classification group that extracts classifications that satisfy the changed conditions,
    The classification hierarchy re-creation system according to any one of claims 1 to 7, wherein the classification hierarchy re-creation unit re-creates a classification hierarchy based on the changed condition.
  9. The re-updating means changes the condition when at least one of the number of classifications in the classification hierarchy, the depth of the classification hierarchy, the number of re-updates of the classification hierarchy, and the presence / absence of a stop instruction does not satisfy a predetermined requirement. The classification hierarchy recreating system according to claim 8.
  10. The clustering means creates a classification group by extracting, from the classifications corresponding to each data in the cluster, a classification in which the number of data belonging to the classification is larger than a predetermined number. The classification hierarchy re-creation system according to any one of the above.
  11. The clustering means of the data processing device clusters data groups associated with the hierarchical classification,
    The clustering means creates a classification group that is a group in which classifications satisfying a predetermined condition are extracted from the classifications corresponding to each data in the cluster,
    The co-occurrence degree calculation means of the data processing device calculates the co-occurrence degree of two classifications selected from the classification group,
    A classification hierarchy re-creation method, wherein a classification hierarchy re-creation unit of the data processing device re-creates the classification hierarchy based on the classification group and the co-occurrence degree.
  12. The classification hierarchy recreating method according to claim 11, wherein the clustering means creates a classification group obtained by dividing the classification group when the classification within the created classification group is more than a predetermined distance.
  13. The co-occurrence degree calculation means calculates the co-occurrence degree based on the co-occurrence frequency that is the number of data in which two classifications co-occur and the number of data belonging to each classification,
    The classification hierarchy recreating means determines whether the two classifications are inclusive relations or consent relations based on the co-occurrence degree,
    The classification hierarchy recreating method according to claim 11 or 12, wherein the classification hierarchy recreating means recreates a classification hierarchy based on a determination result indicating whether two classifications are inclusive relations or consent relations.
  14. If the classification hierarchy recreating means adds a hierarchy in which the inclusion classification is the parent classification and the inclusion classification is the child classification, when the relationship between the two classifications is an inclusion relation, If the relationship between the two categories is a consensus relationship, the classification hierarchy is re-created by creating one category that integrates a small number of the two categories into a category with a large amount of data. The classification hierarchy re-creation method according to claim 13.
  15. If the classification hierarchy re-creation means adds a hierarchy whose child classification is included, the classification hierarchy is re-created by deleting the parent-child relationship of the child classification before the classification hierarchy is re-created. The classification hierarchy recreating method according to claim 14.
  16. The classification hierarchy recreating means recreates the classification hierarchy by deleting the classification when the classification with no data belonging to is a classification having no child classification, and the classification classification without the data belonging to it is set to 1 In the case of a class having only one class, the class is deleted, and the class hierarchy is recreated by creating a hierarchical relationship between the parent class of the class to be deleted and the child class. The classification hierarchy recreating method according to claim 15.
  17. Based on the structured data that is the structured data and the structural part name that is the name for identifying each part of the structured data, the clustering means extracts the part corresponding to the structural part name from the structured data. The classification hierarchy recreating method according to claim 11, wherein the structured data group is clustered using the extracted data.
  18. When the re-updating means of the data processing device does not satisfy the predetermined requirement, the conditions for creating the classification group and the conditions for the co-occurrence degree for re-creating the classification hierarchy Change at least one of the conditions and instruct to update the re-created classification hierarchy again,
    The clustering means creates a classification group that extracts classifications that meet the changed conditions,
    The classification hierarchy re-creation method according to any one of claims 11 to 17, wherein the classification hierarchy re-creation unit re-creates a classification hierarchy based on the changed condition.
  19. The condition is changed when at least one of the number of classification hierarchies, the depth of the classification hierarchies, the number of times of renewal of the classification hierarchies, and the presence / absence of stop instruction does not satisfy the predetermined requirement. The classification hierarchy recreating method according to claim 18.
  20. The clustering means creates a classification group by extracting a classification in which the number of data belonging to the classification is larger than a predetermined number from the classification corresponding to each data in the cluster. The classification hierarchy re-creation method according to any one of the above.
  21. On the computer,
    A clustering process that clusters data groups associated with hierarchical classifications and creates a classification group that is a group in which classifications satisfying predetermined conditions are extracted from the classifications corresponding to each data in the cluster. ,
    A co-occurrence degree calculation process for calculating the co-occurrence degree of two classifications selected from the classification group; and
    A classification hierarchy re-creation program for executing a classification hierarchy re-creation process for re-creating the classification hierarchy based on the classification group and the co-occurrence degree.
  22. On the computer,
    The classification hierarchy re-creation program according to claim 21, wherein when the classification in the created classification group is more than a predetermined distance in the clustering process, a classification group is generated by dividing the classification group.
JP2011521779A 2009-07-06 2010-04-20 Classification hierarchy re-creation system, classification hierarchy re-creation method, and classification hierarchy re-creation program Active JP5621773B2 (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2009160071 2009-07-06
JP2009160071 2009-07-06
PCT/JP2010/002855 WO2011004529A1 (en) 2009-07-06 2010-04-20 Classification hierarchy re-creation system, classification hierarchy re-creation method, and classification hierarchy re-creation program

Publications (2)

Publication Number Publication Date
JPWO2011004529A1 JPWO2011004529A1 (en) 2012-12-13
JP5621773B2 true JP5621773B2 (en) 2014-11-12

Family

ID=43428962

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2011521779A Active JP5621773B2 (en) 2009-07-06 2010-04-20 Classification hierarchy re-creation system, classification hierarchy re-creation method, and classification hierarchy re-creation program

Country Status (3)

Country Link
US (1) US8732173B2 (en)
JP (1) JP5621773B2 (en)
WO (1) WO2011004529A1 (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6043067B2 (en) * 2011-06-03 2016-12-14 株式会社東芝 Stock trading knowledge extraction device and program
JP5308593B2 (en) * 2011-07-25 2013-10-09 楽天株式会社 Genre generator
US8650198B2 (en) * 2011-08-15 2014-02-11 Lockheed Martin Corporation Systems and methods for facilitating the gathering of open source intelligence
JP5583107B2 (en) * 2011-12-02 2014-09-03 日本電信電話株式会社 Keyword place name pair extraction apparatus, method, and program
US9286391B1 (en) * 2012-03-19 2016-03-15 Amazon Technologies, Inc. Clustering and recommending items based upon keyword analysis
EP2857985A4 (en) 2012-05-31 2016-08-03 Toshiba Kk Knowledge extraction device, knowledge updating device, and program
JP6201779B2 (en) * 2014-01-20 2017-09-27 富士ゼロックス株式会社 Information processing apparatus and information processing program
US9928232B2 (en) 2015-02-27 2018-03-27 Microsoft Technology Licensing, Llc Topically aware word suggestions
JP6359045B2 (en) * 2016-03-16 2018-07-18 ヤフー株式会社 Information processing apparatus, information processing method, and information processing program

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH1196177A (en) * 1997-09-22 1999-04-09 Nippon Telegr & Teleph Corp <Ntt> Method for generating term dictionary, and storage medium recording term dictionary generation program
JP2000010996A (en) * 1998-06-24 2000-01-14 Fujitsu Ltd Document arranging device and method therefor
JP2003140942A (en) * 2001-10-19 2003-05-16 Xerox Corp Methods, systems, and articles of manufacture for soft hierarchical clustering of co-occurring objects
US20040181554A1 (en) * 1998-06-25 2004-09-16 Heckerman David E. Apparatus and accompanying methods for visualizing clusters of data and hierarchical cluster classifications
JP2005266866A (en) * 2004-03-16 2005-09-29 Fuji Xerox Co Ltd Document classifying device and classification system generating device and method for document classifying device
JP2008299382A (en) * 2007-05-29 2008-12-11 Fujitsu Ltd Data division program, recording medium with the program recorded thereon, data distribution device and data distribution method

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8055597B2 (en) * 2006-05-16 2011-11-08 Sony Corporation Method and system for subspace bounded recursive clustering of categorical data
JP4884871B2 (en) 2006-07-27 2012-02-29 東京エレクトロン株式会社 Coating method and coating apparatus
US20080109454A1 (en) * 2006-11-03 2008-05-08 Willse Alan R Text analysis techniques

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH1196177A (en) * 1997-09-22 1999-04-09 Nippon Telegr & Teleph Corp <Ntt> Method for generating term dictionary, and storage medium recording term dictionary generation program
JP2000010996A (en) * 1998-06-24 2000-01-14 Fujitsu Ltd Document arranging device and method therefor
US20040181554A1 (en) * 1998-06-25 2004-09-16 Heckerman David E. Apparatus and accompanying methods for visualizing clusters of data and hierarchical cluster classifications
JP2003140942A (en) * 2001-10-19 2003-05-16 Xerox Corp Methods, systems, and articles of manufacture for soft hierarchical clustering of co-occurring objects
JP2005266866A (en) * 2004-03-16 2005-09-29 Fuji Xerox Co Ltd Document classifying device and classification system generating device and method for document classifying device
JP2008299382A (en) * 2007-05-29 2008-12-11 Fujitsu Ltd Data division program, recording medium with the program recorded thereon, data distribution device and data distribution method

Also Published As

Publication number Publication date
WO2011004529A1 (en) 2011-01-13
JPWO2011004529A1 (en) 2012-12-13
US8732173B2 (en) 2014-05-20
US20120109963A1 (en) 2012-05-03

Similar Documents

Publication Publication Date Title
Choo et al. Utopian: User-driven topic modeling based on interactive nonnegative matrix factorization
Treeratpituk et al. Disambiguating authors in academic publications using random forests
Carpineto et al. Exploiting the potential of concept lattices for information retrieval with CREDO.
US7797265B2 (en) Document clustering that applies a locality sensitive hashing function to a feature vector to obtain a limited set of candidate clusters
Boyack et al. Mapping the backbone of science
US8935291B2 (en) System and method for automatically generating systematic reviews of a scientific field
US7266548B2 (en) Automated taxonomy generation
JP3665480B2 (en) Document organize apparatus and method
Chakrabarti et al. A graph-theoretic approach to webpage segmentation
US20080249999A1 (en) Interactive cleaning for automatic document clustering and categorization
CN101408886B (en) Selecting tags for a document by analyzing paragraphs of the document
US20060095852A1 (en) Information storage and retrieval
US7720848B2 (en) Hierarchical clustering with real-time updating
JP5448105B2 (en) Method for retrieving document data from search keywords, computer system and computer program
US7519582B2 (en) System and method for performing a high-level multi-dimensional query on a multi-structural database
US20020133483A1 (en) Systems and methods for computer based searching for relevant texts
US8108392B2 (en) Identifying clusters of words according to word affinities
US8280886B2 (en) Determining candidate terms related to terms of a query
US20090043797A1 (en) System And Methods For Clustering Large Database of Documents
JP5284990B2 (en) Processing method for time series analysis of keywords, processing system and computer program
JP5391633B2 (en) Term recommendation to define the ontology space
US9424294B2 (en) Method for facet searching and search suggestions
GB2395808A (en) Information retrieval
JP2000285140A (en) Device and method for processing document, device and method for classifying document, and computer readable recording medium recorded with program for allowing computer to execute these methods
JP2009258794A (en) Information retrieval program, information retrieval device and information retrieval method

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20130306

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20140401

A521 Written amendment

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20140515

TRDD Decision of grant or rejection written
A01 Written decision to grant a patent or to grant a registration (utility model)

Free format text: JAPANESE INTERMEDIATE CODE: A01

Effective date: 20140826

A61 First payment of annual fees (during grant procedure)

Free format text: JAPANESE INTERMEDIATE CODE: A61

Effective date: 20140908

R150 Certificate of patent or registration of utility model

Ref document number: 5621773

Country of ref document: JP

Free format text: JAPANESE INTERMEDIATE CODE: R150